More than four decades of development has led to the Internet exerting significant influence across everyday consumption. Internet applications are now expanding from consumption to the production field, deeply integrating with real economy sectors such as manufacturing and energy. This poses higher requirements on the real-timeliness, security and reliability, service level classification, massive data processing, and resource scheduling of network communication. In this context, the sustainable development of networks has gradually become the focus of global attention. The fundamental transformation following industry digitalization and intelligentization calls for multi-level and multi-dimensional research on new basic theories and technical methods. These include designing a new network architecture free of existing Internet defects, exploring key technologies suitable for future network application innovation, developing core devices and systems for future networks, and performing large-scale networking verification.
At present, research has been conducted on future network architecture and key technologies both inside and outside China, and Software-Defined Networking (SDN) has attracted worldwide attention. Thanks to the centralized control mechanism, SDN can greatly improve the controllability, manageability, and flexibility of existing networks while effectively reducing Capital Expenditure (CAPEX) and Operating Expenditure (OPEX) of network service providers. In addition, SDN is capable of collecting an extensive amount of network data on the data plane in real time.
However, new application scenarios such as the Internet of Things (IoT) will render human-compiled centralized control programs — such as SDN — incapable of effectively dealing with future network complexity and unexpected network events.
Big data — collected by mechanisms such as network telemetry — can be analyzed by big data analytics and AI technologies in real time. This way, computers with AI capabilities may be able to detect 90 percent of network faults and security risks and suggest solutions to fix them. As a result, experts could solely focus on the remaining 10 percent of problems that machines cannot solve. Machines can also acquire increasingly stronger capabilities to handle complex network problems through continuously iterative AI training. In terms of path planning and traffic scheduling, future networks need to meet the high throughput and low latency requirements of applications. It is very difficult to provide an optimal traffic scheduling solution in real time based on the dynamic status of link loads if traditional path planning algorithms are used. Nevertheless, AI technologies are expected to make proactive prediction and effective scheduling of link traffic possible by using the massive amount of historical traffic data accumulated. AI has great potential to improve network management, fault detection, network security, path planning, traffic scheduling, and many other operations. The intelligence of the “network brain” is becoming a key enabler to address the continuous increase of network scale and complexity.
Introducing new concepts such as AI, to future networks brings technical challenges, one especially worth mentioning is how to improve the reliability of AI decision-making. Path planning and traffic scheduling on networks differ greatly from consumer-oriented services such as voice recognition. The AI-trained deep learning model can allow for certain errors during speech recognition, but errors are usually not permitted on networks. Path planning errors may lead to large-scale network breakdown, causing great loss.
This means, the introduction of AI technologies to future networks must be carried out in phases. The first breakthrough should be using AI technologies to implement network fault detection and network security diagnosis based on big data analytics. In this phase, AI will independently solve simple network problems and merely assist human experts with the more complex issues. As AI gradually matures and the network brain becomes more reliable, proactive planning of network paths will be conducted by the brain, which delivers an optimization efficiency of highly dynamic networks better than that achieved using traditional network algorithms.
We believe that AI technologies will revolutionize the network industry as they mature, providing an unprecedented opportunity for China to build self-sufficient, controllable, and secure new networks.
[This article, originally published in the Chinese Association for Artificial Intelligence Communications, is now republished after being revised by the author.]